Generation of a dense point cloud of a physical object
A dense point cloud (DPC) is generated for a physical object. 2D digital images of the physical object are obtained. Object extractors are defined. A sparse point cloud (SPC) is created from the 2D digital images. A modified SPC is obtained by first applying the object extractors to the 2D digital images to exclude pixels from the 2D digital images and then excluding the datapoints in the SPC that correspond to the pixels that have been excluded from the 2D digital images. The datapoints of the SPC that do not represent the object are excluded from being included in the modified SPC. The DPC is generated by integrating information obtained from the pixels in the 2D digital images with the modified SPC.
1 . A method for generating a dense point cloud (DPC) of a physical object, the method being performed by an image processing device, the method comprising:
obtaining two-dimensional (2D) digital images of the physical object, wherein each 2D digital image is comprised of pixels, and wherein some, but not all of the pixels in each of the 2D digital images, represent the physical object;
defining object extractors, wherein the object extractors are defined to separate a first set of the pixels in the 2D digital images representing the physical object from a second set of the pixels in the 2D digital images not representing the physical object;
creating a sparse point cloud (SPC) from the 2D digital images, wherein the SPC is comprised of datapoints, and wherein some, but not all of the datapoints, represent the physical object;
obtaining a modified SPC from the SPC by mapping the datapoints of the SPC to pixels in the 2D digital images, applying the object extractors to the 2D digital images to exclude pixels from the 2D digital images, and excluding the mapped datapoints in the SPC that correspond to the pixels that have been excluded from the 2D digital images, whereby the datapoints in the SPC that do not represent the object are excluded from being included in the modified SPC; and
generating the DPC by integrating information obtained from the pixels in the 2D digital images with the modified SPC.
2 . The method according to claim 1 , wherein each datapoint has a value, and wherein the values of the datapoints of the SPC that are excluded are nulled.
3 . The method according to claim 1 , wherein obtaining the modified SPC further comprises:
associating each 2D digital image with a respective one of the object extractors;
obtaining modified 2D digital images by applying each object extractor to its associated 2D digital image, whereby the pixels of the associated 2D digital images that do not represent the physical object are excluded; and
obtaining the modified SPC from the modified 2D digital images by excluding the mapped datapoints in the SPC that correspond to the pixels having been excluded from the 2D digital images.
4 . The method according to claim 3 , wherein each pixel has a value, and wherein the values of the pixels of the modified 2D digital images that are excluded are nulled.
5 . The method according to claim 1 , wherein the object extractors are defined by depth maps created from the 2D digital images, and wherein the pixels in the 2D digital images representing the physical object are separated from pixels in the 2D digital images not representing the physical object by a threshold depth value in the depth maps.
6 . The method according to claim 1 , wherein the object extractors are defined by configuring a machine learning training procedure.
7 . The method according to claim 1 , wherein each object extractor is represented by a respective binary mask.
8 . The method according to claim 1 , wherein each of the 2D digital images comprises exchangeable image file format (EXIF) information, and wherein the DPC is generated based on the EXIF information.
9 . The method according to claim 1 , wherein the 2D digital images are captured from an image capturing unit mounted on an unmanned aerial vehicle, UAV.
10 . The method according to claim 1 , wherein the physical object is a piece of telecommunications equipment, a part of a cell site, or a complete cell site.
11 . The method according to claim 1 , wherein the physical object is a building, or part of a building.
12 . An image processing device for generating a dense point cloud (DPC) of a physical object, the image processing device comprising processing circuitry, the processing circuitry being configured to cause the image processing device to:
obtain two-dimensional (2D) digital images of the physical object, wherein each 2D digital image is comprised of pixels, and wherein some, but not all of the pixels in each of the 2D digital images, represent the physical object;
define object extractors, wherein the object extractors are defined to separate a first set of the pixels in the 2D digital images representing the physical object from a second set of the pixels in the 2D digital images not representing the physical object;
create a sparse point cloud (SPC) from the 2D digital images, wherein the SPC is comprised of datapoints, and wherein some, but not all of the datapoints, represent the physical object;
obtain a modified SPC by mapping the datapoints of the SPC to pixels in the 2D digital images, applying the object extractors to the 2D digital images to exclude pixels from the 2D digital images, and excluding the mapped datapoints in the SPC that correspond to the pixels that have been excluded from the 2D digital images, whereby the datapoints of the SPC that do not represent the object are excluded from being included in the modified SPC; and
generate the DPC by integrating information obtained from the pixels in the 2D digital images with the modified SPC.
13 . A computer program product for generating a dense point cloud (DPC) of a physical object, the computer program product comprising a non-transitory computer readable medium storing computer code which, when run on processing circuitry of an image processing device, causes the image processing device to:
obtain two-dimensional (2D) digital images of the physical object, wherein each 2D digital image is comprised of pixels, and wherein some, but not all of the pixels in each of the 2D digital images, represent the physical object;
define object extractors, wherein the object extractors are defined to separate a first set of the pixels in the 2D digital images representing the physical object from a second set of the pixels in the 2D digital images not representing the physical object;
create a sparse point cloud (SPC) from the 2D digital images, wherein the SPC is comprised of datapoints, and wherein some, but not all of the datapoints, represent the physical object;
obtain a modified SPC by mapping the datapoints of the SPC to pixels in the 2D digital images, applying the object extractors to the 2D digital images to exclude pixels from the 2D digital images, and excluding the mapped datapoints in the SPC that correspond to the pixels that have been excluded from the 2D digital images, whereby the datapoints of the SPC that do not represent the object are excluded from being included in the modified SPC; and
generate the DPC by integrating information obtained from the pixels in the 2D digital images with the modified SPC.